Passive-Aggressive online learning with nonlinear embeddings
作者：
Highlights：
• A combination of Passive-Aggressive algorithm and the Max-Out function is proposed.
• This approach is an alternative to Kernel-based methods in online learning problems.
• The algorithm does not rely on budget strategies over the set of support vectors.
• The model is fast and uses nonlinear learned embeddings, avoiding support vectors.
• This approach has obtained better results in commonly used online learn- ing benchmarks.
摘要
•A combination of Passive-Aggressive algorithm and the Max-Out function is proposed.•This approach is an alternative to Kernel-based methods in online learning problems.•The algorithm does not rely on budget strategies over the set of support vectors.•The model is fast and uses nonlinear learned embeddings, avoiding support vectors.•This approach has obtained better results in commonly used online learn- ing benchmarks.
论文关键词：Online learning,Nonlinear functions,Passive-Aggressive,Binary classification,Nonlinear embedding
论文评审过程：Received 3 April 2017, Revised 19 January 2018, Accepted 24 January 2018, Available online 31 January 2018, Version of Record 23 February 2018.
论文官网地址：https://doi.org/10.1016/j.patcog.2018.01.019
